Inequality in Health Outcomes in India: The Role of Caste and Religion Vani K. Borooah* University of Ulster September 2007 Abstract The “social gradient to health” - whereby people belonging to groups higher up the social ladder had better health outcomes than those belonging to groups further down - is essentially a Western construct; there has been very little investigation into whether, in developing countries also, people’s state of health is dependent on their social status. The purpose of this paper is to evaluate the relative strengths of economic and social status in determining the health status of persons in India. In other words, even after controlling for non-community factors, did the fact that Indians belonged to different social groups, encapsulating different degrees of social status, exercise a significant influence on the state of their health? The existence of a social group effect would suggest that there was a “social gradient” to health outcomes in India. Furthermore, there was the possibility that the “social gradient” existed with respect to some outcomes but not to others. In investigating this, the paper addresses, in the Indian context, an issue which les at the heart of social epidemiology: estimating the relative strengths of individual and social factors in determining health outcomes. JEL: D53 I12,O15 Keywords: Health outcomes, Caste, Religion, India

*

School of Economics and Politics, University of Ulster, Newtownabbey, Northern Ireland BT37 0QB. E-mail: [email protected]

1. Introduction The publication of the Black report (Black et. al., 1980) spawned a number of studies in industrialised countries which examined the social factors underlying health outcomes. The fundamental finding from these studies, particularly with respect to mortality and life expectancy, was the existence of “a social gradient” in mortality: “wherever you stand on the social ladder, your chances of an earlier death are higher than it is for your betters” (Epstein, 1998). The social gradient in mortality was observed for most of the major causes of death: for example, Marmot (2000) showed that, for every one of twelve diseases, the ratio of deaths (from the disease) to numbers in a Civil Service grade rose steadily as one moved down the hierarchy. Since, in the end, it is the individual who falls ill, it is tempting for epidemiologists to focus on the risks inherent in individual behaviour: for example, smoking, diet, and exercise. However, the most important implication of a social gradient to health outcomes is that people’s susceptibility to disease depends on more than just their individual behaviour; crucially, it depends on the social environment within which they lead their life (Marmot, 2000 and 2004). Consequently, the focus on inter-personal differences in risk might be usefully complemented by examining differences in risk between different social environments. For example, even after controlling for inter-personal differences, mortality risks might differ by occupational class. This might be due to the fact that while low status jobs make fewer mental demands, they cause more psychological distress than high status jobs (Karasek and Marmot, 1996; Griffin et. al., 2002; Marmot, 2004) with the result that people in higher level jobs report significantly less job-related depression than people in lower-level jobs (Birdi et.al., 1995).

1

In turn, anxiety and stress are related to disease: the stress hormones that anxiety releases affect the cardiovascular and immune systems with the result that prolonged exposure to stress is likely to inflict multiple costs on health in the form of inter alia increased susceptibility to diabetes, high blood pressure, heart attack, and stroke (Marmot, 1986; Wilkinson and Marmot, 1998; Brunner and Marmot, 1999). So, the social gradient in mortality may have a psychosocial basis, relating to the degree of control that individuals have over their lives.1 The “social gradient to health” is essentially a Western construct and there has been very little investigation into whether, in developing countries as well, people’s state of health is dependent on their social status. For example, in India, which is the country studied in this paper, we know from studies of specific geographical areas that health outcomes differ systematically by gender and economic class (Sen, Iyer, and George, 2007). In addition, local government spending on public goods, including health-related goods, is, after controlling for a variety of factors, lower in areas with greater caste fragmentation compared to ethnically more homogenous areas (Sengupta and Sarkar, 2007). Considering India in its entirety, two of its most socially depressed groups the Adivasis2 and the Dalits3 - have some of the worst health outcomes: for example, as Guha (2007) observes, 28.9 percent of Adivasis and 15.6 percent of Dalits have no access to doctors or clinics and only 42.2 percent of Adivasi children and 57.6 percent

1

Psychologists distinguish between stress caused by a high demand on one’s capacities – for example, tight deadlines – and stress engendered by a low sense of control over one’s life. 2 There are about 85 million Indians classified as belonging to the “Scheduled Tribes”; of these, Adivasis (meaning original inhabitants”) refer to the 70 million who live in the heart of India, in a relatively contiguous hill and forest belt extending across the states of Gujarat, Rajasthan, Maharashtra, Madhya Pradesh, Chhattisgargh, Jharkhand, Andhra Pradesh, Orissa, Bihar, and West Bengal (Guha, 2007). 3 Dalits, who number about 18 million, refer to those who belong India’s “Scheduled Castes” and may be broadly identified with the “untouchable” castes i.e. those with whom physical contact – most usually taken to be the acceptance of food or water – is regarded by upper-caste Hindus as ritually polluting or unclean

2

of Dalit children have been immunised. Of course, it is possible that the relative poor health outcomes of India’s socially backward groups has less to do with their low social status and much more to do with their weak economic position and with their poor living conditions. The purpose of this paper is precisely to evaluate the relative strengths of economic and social status in determining the health status of persons in India. In other words, even after controlling for non-community factors, did the fact that Indians belonged to different social groups, encapsulating different degrees of social status, exercise a significant influence on the state of their health? We answer this question using data from the Morbidity and Health Care Survey (hereafter, referred to as the M&HC Survey), for the period January-June 2004, conducted over all the states and union territories in India, by the Government of India’s National Sample Survey Organisation (NSSO).4 The M&HC Survey covered 73,868 households, encompassing 383,338 individuals. It examined several aspects of morbidity and health care of the respondents but, from this study’s perspective, three of these are of note: (i)

Particulars of household members who died with the past 365 days.

(ii)

Particulars of economic independence and ailments on the date of survey of persons aged 60 years or more (hereafter, “elderly” persons).

(iii)

Particulars of prenatal and postnatal care for ever married women.

These aspects of morbidity and health care could inter alia be correlated with the social background of the households to which the respondents belonged. The M&HC Survey offered information about households in terms of the following social groups: 1. Adivasis (see note 2)

4

For background on the NSSO see Tendulkar (2007).

3

2. Scheduled Tribe (ST) Christians.5 3. Dalits (see note 3). 4. Non-Muslims from the Other Backward Classes (OBC).6 5.

Muslims from the OBC.

6.

Muslims not from the OBC.

7. Forward Caste Hindus (hereafter, simply “Hindus”).7 8. Non-ST Christians. 9. Sikhs. 10. Other Religions. The primary aim of this paper is to examine whether the following health outcomes varied systematically according to the social group to which people belonged: (i)

The age of death

(ii)

The self-assessed health status of persons 60 years of age or more

(iii)

The likelihood of elderly persons, who were in poor health, taking treatment for their ailments.

(iv)

The likelihood of women receiving prenatal and postnatal treatment

The purpose was to investigate whether, after controlling for several nongroup factors that might impinge on health outcomes, people’s health outcomes were significantly affected by their social group. The existence of a social group effect – whereby groups higher up the social ladder had better health outcomes than groups 5

As Guha (2007) notes, Scheduled Tribe Christians have been exposed to modern education in English and have a much greater chance of being absorbed in the modern economy. They also live mainly in the hills of North-East India which are some of the remotest and less accessible parts of the country. 6 These are persons who, while not belonging to the Scheduled Tribes or Scheduled Castes, nevertheless belong to economic and socially backward groups. 7 Forward caste Hindus were Hindus who were not included in the OBC/Dalit/ST categories. However, since the designation of groups in the OBC category is a state responsibility a particular (caste) group may be included in the OBC category in one state (i.e. be excluded from forward caste Hindus) but be excluded from the OBC category in another state (i.e. be included in forward caste Hindus).

4

further down – would suggest that there was a “social gradient” to health outcomes in India. Furthermore, there was the possibility that the “social gradient” existed with respect to some outcomes but not to others. In so doing, the paper addresses, in the Indian context, an issue which les at the heart of social epidemiology: estimating the relative strengths of individual and social factors in determining health outcomes.

2. Deaths in Households The M&HC Survey asked households if any of their members had died in the previous year and, if the answer was in the affirmative, collected information about the deceased and some of the circumstances surrounding the deaths. In total, 1,716 deaths were reported: 1,634 of these deaths (95 percent) were from households which had experienced a single death in the past year; 70 deaths (4 percent) occurred in households which had experienced two deaths; and 12 deaths (1 percent) occurred in households which had experienced three deaths. Of these 1,716 deaths, 9.1 percent were Adivasis, 17.6 percent were Dalits, and 12 percent were Muslim, and 21.3 were Hindus (Table 1). By contrast, Adivasis, Dalits, and Hindus comprised 7.9, 16.9, and 23.6 percent, respectively, of the total of the 383,288, persons in the M&HC-NSS sample. Thus, in respect of Adivasis and Dalits, there was a difference between their proportionate presence in the number of deaths and their proportionate presence in the sample.8 A more marked difference between the groups was in terms of the mean and median ages at death: as Table 1 shows, the mean age of death was 43.3 years for Adivasis, 41.6 years for Dalits, 43.4 years for OBC Muslims, and 43.8 years for nonOBC Muslims; by contrast, the mean age at death was 57.5 for Sikhs and non-ST

8

Of the 1,716 deceased, 58 percent were men. For all the groups the majority of deaths were male except for ST Christians where 55 percent of the 65 deaths in this group were female.

5

Christians, 54.2 years for Hindus, 49.4 years for ST Christians, and 48.4 years for the non-Muslim OBC. Table 2 shows whether the deceased received medical attention before death. The group least likely to receive medical attention before death were Adivasis and ST Christians: only 59 percent of Adivasi deaths and 53 percent of ST Christian deaths received medical attention in contrast to 76 percent of Dalit deaths and 73 percent of Muslim deaths. Although, in terms of the overall sample, there was little difference between the proportions of men and women receiving medical attention before death (69 percent men, 71 percent women) there were marked gender differences between some of the social groups: Muslim deaths were more likely to receive medical attention if they were women (80 percent against 70 percent for on-OBC Muslims) while Dalit deaths were more likely to receive medical attention if they were men (80 percent against 71 percent). Table 3 presents the estimates from regressing the “age at death” on a number of explanatory variables.9 The first column shows the regression estimates obtained from all deaths in the sample; the second and third columns show the regression estimates obtained from all deaths in, respectively, the “forward” and “backward” states (and union territories) of India.10 The mean ages at death in the forward and backward states were, respectively, 52.4 and 43.7 years – a difference of 8.7 years. After imposing all the controls shown in Table 3, the difference between forward and backward states in their averages at death was reduced to 7.4 years (Table 3, column 1)

9

Excluding the 27 deaths which occurred during pregnancy. “Forward states”: Andhra Pradesh, Chandigarh, Dadra and Nagar Haveli, Daman and Diu, Delhi, Goa, Gujarat, Haryana, Himachal Pradesh, Karnataka, Kerala, Maharashtra, Pondicherry, Punjab, Tamil Nadu, West Bengal. The remaining states and union territories were classed as “backward” states. 10

6

The second variable in the regression was gender: Table 3 shows that, after controlling for other variables, there was no significant difference between the average ages of the male and female deceased.11 The next variable was whether the household type in which the deceased lived was a “labourer” household12: Table 3 shows that, after imposing all controls, the average age at death was 4.3 years lower for labourer, compared to non-labourer, households. Since the sample differences between non-labourer and labourer households in the ages of their deceased was 6.9 years, imposing the controls, reduced this difference but without eliminating it. Table 3 also shows that the average age at death was significantly different between labourer and non-labourer households in the forward states but not in the backward states. The average age of the deceased was significantly higher, by 4.9 years, in rural, compared to urban, areas and, in the backward states, the rural-urban difference average age at death was 7.8 years; however, in the forward states, there was no significant difference between rural and urban areas in the average age at death. After these four controls – state type, gender, household type, and rural-urban sectors – the next set of controls related to the conditions in which the deceased lived. 1.

The first component of this was the type of housing structure in which the deceased lived: this variable (“structure”) was assigned the value 1 if the type was pucca, or semi-pucca, or “serviceable” kutcha (i.e. good); and 0 otherwise.

2.

The second component of living conditions related to the quality of the latrines used by the deceased: the variable “latrine” was assigned the value 1 if the latrines were flushing toilets or emptied into a sceptic tank; and 0 otherwise.

11

The sample averages for age at death were 48.4 and 46.2 years for male and female deaths, respectively. 12 Agricultural or other labour for rural households and casual labour for urban households.

7

3.

The third component of living conditions related to the quality of the drains: the variable “drain” was assigned the value 1 if the drains associated with the deceased’s home were underground or were covered pucca; and 0 otherwise.

4.

The fourth component of living conditions related to the quality of the source of drinking water used by the deceased: the variable “water source” was assigned the value 1 if the source of drinking water was from a tap; and 0 otherwise.

5.

The fifth component of living conditions related to whether the drinking water used by the deceased was treated: the variable “water treated” was assigned the value 1 if the drinking water was treated; and 0 otherwise.

6.

If the drinking water in the deceased’s household was treated, the sixth component of living conditions related to the nature of the treatment of the drinking water: the variable “water treatment” was assigned the value 1 if the nature of treatment was boiling, filtering, or ultra-violet/resin/reverse osmosis; and 0 otherwise.

7.

The seventh, and last, component of living conditions related to the nature of the cooking fuel used by the deceased’s household: the variable “cooking fuel” was assigned the value 1 if the cooking fuel was gas, gobar gas, kerosene, or electricity; and 0 otherwise.

Table 3 shows that, of these seven living conditions controls, it was only the nature of treatment of drinking water and of the type of cooking fuel used that had a significant effect on the age of the deceased. The average age of deceased persons whose drinking water was boiled or treated through chemical means was, over India

8

in its entirety, 11.8 years higher than that of those whose drinking water was either not treated or treated through “other means”; for “forward” and “backward” states, this difference was, respectively, 10.9 and 12.6 years. Similarly, the average age of deceased persons whose households used gas (including gobar gas), kerosene, or electricity as their cooking fuel was 5 years higher than that of those whose households used “other” fuels.13 After controlling for the living conditions of the deceased, the next set of controls related to the economic position of the deceased’s households. This was measured by a household’s consumer expenditure in the past 30 days. Table 3 shows that an increase of Rs.1,000 in monthly household expenditure would raise the average age of death by approximately 0.4 years though, it must be added that, after the other controls had been imposed, the significance of the relation between monthly expenditure and the mean age of death was very weak. Table 3 shows that, even after imposing all the above controls, the average age of the deceased was significantly affected by the social group to which they belonged. Compared to the average age at death of Hindus (the control group), the average age at death of: Adivasis was 4.9 years lower for India in its entirety and 6.7 years lower for the backward states; Dalits was 7.1 years lower for India in its entirety and 11.5 years lower for the backward states; OBC Muslims was 8.6 years lower for India in its entirety and 11.9 years lower for the backward states; non-OBC Muslims was 6.1 years lower for India in its entirety and 6.8 years lower for the forward states. By contrast, there was no significant difference in the ages of deceased persons between Hindus and the (non-Muslim) OBCs.

13

For backward states, this difference was significant only at 10% and for forward states it was not even significant at this level.

9

3. The Health of Elderly Persons Table 4 shows the perceptions of persons, aged 60 years or more (“elderly persons”), about their state of health: excellent/very good; good/fair; poor. While 25 percent of the entire sample of 33,155 elderly persons described themselves as being in poor health, this description was offered by 28 percent of Dalits and 31 percent of Muslims (OBC and non-OBC). By contrast, only 16 percent of ST Christians and 20 percent of Adivasis regarded themselves as being in poor health. Table 5 shows the marginal probabilities obtained from estimating an ordered logit model in which the dependent variable took the value 1, 2, or 3 depending on whether a person described his/her state of health as excellent/very good; good/fair; poor. The marginal probability associated with a variable is the change in the probability of an outcome, following a change in the value of a variable. For each variable, these probabilities sum to zero across the three outcomes (i.e. the three states of health) and for discrete variables – as are all the explanatory variables used, except age - the marginal probabilities refer to changes in the probability of the outcomes, consequent on a move from the default category for that variable to the category in question.14 For ease of exposition, the subsequent discussion focuses, in the main, on the marginal probability of regarding oneself to be in poor health (hereafter, simply, “the probability of poor health”). According to Table 5: (i) moving from a backward state to a forward state would reduce the probability of poor health by 1.1 points; (ii) being female would increase the probability of poor health by 4.3 points. The effect of age on the probability of poor health depends not only upon the increase in age but, because of

14

In an ordered logit model, the signs of the coefficient estimates associated with a variable do not predict the directions of change in the probabilities of the outcomes and these probabilities have to be separately calculated.

10

the presence of the non-linear term age2, also upon the age itself. So, for an additional year in age from N years, age2 would increase by (N+1)2 – N2 = 2N+1. Therefore, if N=60, the probability of poor health would increase by 5.2 - 1210.03 = 1.57 points for an additional year; if N=75, the probability of poor health would increase by 5.2 1510.03 = 0.67 points for an additional year. In other words, the probability of poor health would increase with age, but at a diminishing rate, and, after a certain age (N=87), would not change with increasing years. Table 5 suggests that people’s perception of the state of their health was significantly affected by their level of education. Compared to an illiterate person (the default level), the probability of poor health was: 1.6 points lower for a person educated up to primary schooling (“low education”); 4.2 points lower for a person educated above primary and up to secondary level; and 7.4 points lower for a person educated up to higher secondary or more. Living conditions exerted a significant effect on the probability of poor health: good housing conditions (“structure”) reduced this probability by 3.2 points; a good source of drinking water (“water source”) reduced it by 3.1 points while treating drinking water and, furthermore, treating it “properly” reduced it by, respectively, 4.3 and 5.4 points; lastly, using a “clean” fuel for cooking lowered the probability of poor health by 3.4 points. In total, therefore, good living conditions were capable of reducing the probability of poor health by nearly 20 points. Over and above, these factors, the economic position of a household also had a significant effect on the probability of poor health: compared to elderly persons from households whose monthly expenditure was in the top quartile (the control group), elderly persons from households whose monthly expenditure was in the lowest,

11

second, and third quartile were more likely to be in poor health by, respectively, 4.4, 3.7, and 2.2 points. Lastly, even after controlling for all the above factors, Table 5 shows that the social groups to which people belonged had a significant effect on their probabilities of poor health: compared to Hindus (the control group), Adivasis and ST Christians were less likely to be in poor health by, respectively, 5.0 and 8.3 points; on the other hand, Dalits, OBC Muslims, and non-OBC Muslims were more likely to be in poor health by, respectively, 2.6, 5.5, and 8.1 points. Table 6 records the primary ailments of elderly persons who regarded their state of health as “poor”. For example, of the 258 such persons who happened to be Adivasis, 5 percent primarily suffered from gastro-intestinal problems (GASTR), 9 percent from cardiovascular disease (CARD), 11 percent from respiratory problems (RESP); 12 percent from disorders of the joints (JOINT); 2 percent from diseases of the kidney or urinary system (KIDNY); 4 percent from neurological disorders (NEURO); 10 percent from eye disorders (EYES); 2 percent from diabetes (DIABT); 4 percent from fever-related illness (FEVER); 24 percent from disabilities (DISAB); 2 percent from accidents/injuries/burns (ACC); 1 percent from cancer (CANC); and 14 percent from other ailments (OTHER). The distribution of the incidence of cardiovascular disease (including hypertension) between the social groups is interesting: 33 percent of non-ST Christians and 19 percent of Hindus, Sikhs, and persons from other religions – aged 60 or more and in poor health – suffered from cardiovascular diseases; by contrast, this ailment affected only 4 percent of ST Christians, 9 percent of Adivasis and Dalits, 12 percent of non-Muslim OBCs, and 16 percent of Muslims. Similarly, compared to the 6 percent of Hindus who were diabetic, only 2 percent of Adivasis and 1 percent

12

of Dalits had diabetes. On the other hand, 33 percent of ST Christians, 24 percent of Adivasis, 16 percent of Dalits , and 18 percent of OBC Muslims – compared to only 12 percent of Hindus – suffered from disabilities.15 Table 7 shows the proportion of elderly persons, who were in poor health, from the different social groups who were not taking any treatment for their ailments: 38 percent of Adivasis, 44 percent of ST Christians, and 33 percent of Dalits, were not taking any treatment for their ailments in contrast to 15 percent of Hindus, 11 percent of non-ST Christians, and 18 percent of Sikhs. In order to determine the probabilities of the different persons taking/not taking treatment for their ailments, we estimated a logit model over the sample of 5,484 elderly persons, who were in poor health, in which the dependent variable took the value 1 if the person was taking treatment and 0 if he/she was not. The marginal probabilities from this model are shown in Table 8. Compared to living in a “backward” state, living in a “forward” state significantly increased the probability of taking treatment by 11.0 points. However, there was no significant difference between women and men, or between persons in the rural and urban sectors, in their probabilities of taking treatment. Having a living daughter had no significant effect on the probability of taking treatment though having a living son raised it by 8.3 points! The level of education of a person, and the economic position of his/her household, had a significant effect on the probability of taking treatment. Compared to an illiterate person (the default level), the probability of taking treatment was: 6.6 points higher for a person educated up to primary schooling (“low education”); 10.8 points higher for a person educated above primary and up to secondary level; and 9.4

15

Locomotor; visual (including blindness, excluding cataract); speech; hearing.

13

points higher for a person educated up to higher secondary or more. Compared to persons from households whose monthly expenditure was in the top quartile (the control group), persons from households whose monthly expenditure was in the lowest, second, and third quartile were less likely to take treatment by, respectively, 16.1, 11.5, and 4.6 points. Another set of factors affecting the probability of people taking treatment comprised their: degree of economic independence, living arrangements, and degree of mobility. Compared to a person who was totally dependent (the default case), the probability of taking treatment was 4.9 points higher for someone who was completely independent and 3.5 points higher someone who was only partially dependent. Compared to living with a spouse, people living without a spouse – whether living alone or with others – were less likely, by 5.1 points, to take treatment. Compared to persons who were totally mobile or else with mobility restricted to the home, people who were confined to bed were more likely, by 5.5 points, to take treatment. However, even after controlling for all the above factors, Table 8 shows that the social groups to which people belonged had a significant effect on their probabilities of taking treatment: compared to Hindus (the control group), Adivasis, ST Christians, and Dalits were less likely to take treatment by, respectively, 10.6, 22.5, and 9.0 points; non-OBC Muslims and the non-Muslim OBCs were less likely to take treatment by, respectively, 7.7 and 5.7 points. Table 9 assesses the predictive performance of the logit model of taking treatment. A person was predicted as taking (not taking) treatment if the predicted probability from the logit model, of his taking treatment, was greater (less) than half. Table 9 shows that of the 5,238 persons predicted to be taking treatment, 4,027 were

14

actually taking treatment – a predictive accuracy of 77 percent; however, only 131 of the 246 predicted to be not taking treatment, were actually not taking treatment – a predictive accuracy of 53 percent. Overall, therefore, 4158 persons out of 5,484 were correctly classified – a predictive accuracy of 76 percent.

4. Prenatal and Postnatal Care The M&HC-NSS provided information, by social group, on the prenatal and postnatal care received by ever married women below 50 years of age. Table 10 shows that, compared to 15 percent of Hindu women who did not receive prenatal care, such care was not received by: 31 percent of Adivasis, 38 percent of ST Christians, 26 percent of Dalits, 33 percent of OBC Muslims, and 26 percent of nonOBC Muslims. Similarly, compared to 27 percent of Hindu women who did not receive postnatal care, such care was not received by: 44 percent of Adivasis and ST Christians, 37 percent of Dalits, 36 percent of OBC Muslims, and 34 percent of nonOBC Muslims. In order to determine the probabilities of women receiving prenatal and postnatal care, we estimated, a logit model in which the dependent variable took the value 1 if the woman received the relevant care and 0 if she did not. The marginal probabilities from this model are shown in Table 11. Compared to living in a “backward” state, living in a “forward” state significantly increased the probability of prenatal care by 15.3 points but it did not have a significant effect on the probability of postnatal care. However, compared to urban women, the probability of rural women receiving prenatal and postnatal care was significantly lower by, respectively, 2.8 and 4.7 points. The level of education of women had a significant effect on the probability of their receiving both prenatal and postnatal care. Compared to an illiterate person (the default level), the probabilities of receiving prenatal and postnatal care were,

15

respectively: 9.0 and 4.0 points higher for a person educated up to primary schooling (“low education”); 14.0 and 11.0 points higher for a person educated above primary and up to secondary level; and 15.7 and 14.0 points higher for a person educated up to higher secondary or more. The economic position of the women’s households exercised a significant positive influence on their probability of receiving prenatal care but not on their probability of receiving postnatal care: compared to women from households whose monthly expenditure was in the top quartile (the control group), women from households whose monthly expenditure was in the lowest, second, and third quartile were less likely to take treatment by, respectively, 3.3, 4.5, and 2.0 points. However, even after controlling for all the above factors, Table 11 shows that the social groups to which women belonged had a significant effect on their probabilities of receiving prenatal care: compared to Hindus (the control group), ST Christians, OBC Muslims, non-OBC Muslims were less likely to receive prenatal care by, respectively, 11.5, 8.8, and 4.3 points and non-ST Christians were more likely to receive prenatal care by 16.3 points. By contrast, after controlling for all the above factors, the effects of social group on the probability of receiving postnatal care were much more muted: the only significant social group effects were that, compared to Hindus, ST Christians were less likely (by 12.3 points), and non-ST Christians were more likely (by 17.3 points), to receive postnatal care.

5. Conclusions This paper investigated whether there was a social gradient to health in India with respect to four health outcomes: the age at death; the self-assessed health status of elderly persons; the likelihood of elderly persons, who were in poor health, taking treatment for their ailments; and the likelihood of receiving prenatal and postnatal

16

care. The evidence suggested that living in a forward state (compared to living in a backward state) and belonging to a relatively affluent household significantly improved all four health outcomes. In addition, the age at death and the self-assessed health status of elderly persons was significantly affected by their household living conditions. The level of education of persons exercised a significant influence on the likelihood of their receiving treatment or care. Ceteris paribus, the likelihood of elderly people, who were in poor health, taking treatment increased with their level of education; similarly, compared to poorly educated women, better educated women were more likely to receive prenatal and postnatal care. However, even after controlling for these “group independent” factors, the social group to which people in India belonged had a significant effect on their health outcomes. Compared to (forward caste) Hindus, the average age at death in India – after imposing all the controls - was 4.9 years lower for Adivasis, 7.1 years lower for Dalits, and 6.1 years lower for Muslims. Similarly, compared to elderly Hindus, elderly Dalits, OBC Muslims, and non-OBC Muslims were – after imposing all the controls - more likely to be in poor health by, respectively, 2.6, 5.5, and 8.1 points. Again, compared to elderly Hindus in poor health, Adivasis, ST Christians, and Dalits were – after imposing all the controls - less likely to take treatment by, respectively, 10.6, 22.5, and 9.0 points and non-OBC Muslims and the non-Muslim OBCs were less likely to take treatment by, respectively, 7.7 and 5.7 points. Lastly, compared to Hindus, ST Christians, OBC Muslims, non-OBC Muslims were – after imposing all the controls - less likely to receive prenatal care by, respectively, 11.5, 8.8, and 4.3 points

17

There can be little doubt, therefore, that, on the basis of data from the M&HC sample, the sample analysed in this paper offered prima facie evidence of a social group bias to health outcomes in India. However, it is important to note that there are several deficiencies inherent in this study. First, there are important health-related attributes of individuals (smoking, diet, taking exercise, the nature of work) which are not - and, indeed, given the limitations of the data, cannot – be taken account of. All these factors are included in the package of factors termed “unobservable”. If these unobservable factors were randomly distributed among the population this, in itself, would not pose a problem. However, there is evidence that there may be a group bias with respect to at least some of these factors. For example, if hard physical work is more inimical to health than more sedentary jobs, then of males aged 25-44 years, 42 percent of Adivasi and 47 percent of Dalits, compared to only 10 percent of Hindus, worked as casual labourers (Borooah et. al. 2007). There is a natural distinction between inequality and inequity in the analysis of health outcomes. Inequality reflects the totality of differences between persons, regardless of the source of these differences and, in particular, regardless of whether or not these sources stem from actions within a person's control. Inequity reflects that part of inequality that is generated by factors outside a person's control. In a fundamental sense, therefore, while inequality may not be seen as “unfair”, inequity is properly regarded as being unfair. The point about group membership is that while it may not be the primary factor behind health inequality, it is the main cause of health inequity. This paper's central message, conditional on the caveats noted earlier, is that being an Adivasi, Dalit, or Muslim in India seriously impaired, using the language of Sen (1992), the capabilities of persons to function in society. This is because, as this study has

18

shown, if you stand at the bottom of the social ladder in India, your risk of suffering premature death, poor health, and a lack of treatment and care is substantially higher than it is for your betters.

References BIRDI, KAMALJIT, WARR, PETER, and OSWALD, ANDREW, 1995. “Age Differences in Three Components of Employee Well-Being,” Applied Psychology, vol. 44, pp. 345-73. BLACK, DOUGLAS, MORRIS, JERRY, SMITH, CYRIL, TOWNSEND, PETER. (1980), Inequalities in Health: A Report of a Research Working Group, London: Department of Health and Social Security. BOROOAH, VANI K., DUBEY, AMARESH, and IYER, SRIYA, 2007. “The Effectiveness of Jobs Reservation: Caste, Religion, and Economic Status in India”, Development & Change, vol. 38, pp. 423-455.. BRUNNER, ERIC and MARMOT, MICHAEL, 1999. “Social Organisation, Stress and Health”, in M. Marmot and R. Wilkinson (eds), The Social Determinants of Health, Oxford University Press: New York, pp. 17-43. EPSTEIN, HELEN, 1998. “Life and Death on the Social ladder”, The New York Review of Books, vol. XLV, pp. 26-30. GRIFFIN, JOAN. M., FUHRER, REBECCA, STANSFELD, STEPHEN A., and MARMOT, MICHAEL, 2002. “The Importance of Low Control at Work and Home on Depression and Anxiety: Do These Effects Vary by Gender and Social Class?”, Social Science and Medicine, vol. 54, pp. 783-98. GUHA, RAMCHANDRA, 2007. “Adivasis, Naxalities, and Indian Democracy”, Economic and Political Weekly, vol. XLII, pp. 3305-3312.

19

KARASEK, ROBERT, and MARMOT, MICHAEL, 1996. “Refining Social Class: Psychosocial Job Factors”, paper presented at The Fourth International Congress of Behavioral Medicine, Washington, D.C., March 13-16. MARMOT, MICHAEL, 1986. “Does Stress Cause Heart Attacks”, Postgraduate Medical Journal, vol. 62, pp. 683-686.MARMOT, MICHAEL, 2000, “Multilevel Approaches to Understanding Social Determinants”, in L. Berkman and I. Kawachi (eds), Social Epidemiology, Oxford University Press: New York, pp. 349367. MARMOT, MICHAEL, 2004. Status Syndrome: How Our Position on the Social Gradient Affects Longevity and Health, London: Bloomsbury Publishing. SEN, GITA, IYER ADITI, and GEORGE, ASHA, 2007. “Systematic Hierarchies and Systemic Failures: Gender and Health Inequalities in Koppal District”, Economic and Political Weekly, vol. XLII, pp. 682-690. SENGUPTA, JHUMUR and SARKAR, DEBNARAYAN, 2007. “Discrimination in Ethnically Fragmented Localities”, Economic and Political Weekly, vol. XLII, pp. 3313-3322. TENDULKAR, SURESH. (2007), “National Sample Surveys” in K. Basu (ed), The Oxford Companion to Economics in India, New Delhi: Oxford University Press (p. 367-370). WILKINSON, RICHARD G. and MARMOT, MICHAEL, 1998. Social Determinants of Health: The Solid Facts, World Health Organisation: Copenhagen.

20

Table 1 Deaths in India by Social Group NSS Persons NSS Deaths by by Social Persons by Social Group Social Group (%) Group (%) Total: 1,716 Adivasi 30,158 7.9 9.2 Christian ST 15,160 4.0 3.8 Dalits 64,942 16.9 17.6 OBC (non-Muslim) 125,508 32.8 33.4 OBC (Muslim) 18,591 4.9 4.8 Hindu (FC) 90,371 23.6 21.3 Muslim (non-OBC) 29,785 7.8 7.2 Christian (non-ST) 3,428 0.9 1.1 Sikh 3,268 0.9 1.2 Other Religion 2,077 0.5 0.5 Total 383,288 100 100 Source: NSS 60th Round, Health File

1

Mean Age of Death by Social Group 43.3 49.4 41.6 48.4 43.4 54.2 43.8 57.6 57.5 64.6 47.7

Median Age of Death by Social Group 45 55 45 55 51 60 50 60 65 70 54

Table 2 Medical Attention Received Before Death by Gender and Social Group Total Medical Total Medical Total Medical Deaths Attention Male Attention Female Attention Received Deaths Received Deaths Received Before Before Before Death as Death as Death as % of % of % of Total Total Total Deaths Male Female Deaths Deaths Adivasi 157 59 85 52 72 67 Christian ST 64 53 28 57 36 50 Dalits 302 76 166 80 136 71 OBC (non573 69 338 66 235 74 Muslim) OBC (Muslim) 82 73 46 70 36 78 Hindu (FC) 366 71 215 72 151 70 Muslim (non123 74 73 70 50 80 OBC) Christian (non18 67 13 62 5 80 ST) Sikh 21 81 14 86 7 71 Other Religion 8 75 7 71 1 100 Total 1,714 70 985 69 729 71 th Source: NSS 60 Round, Health File

2

Table 3: Regression Estimates of the Age at Death Equation, by “Forward” and “Backward” States+ All Deaths Forward State Female Labourer Rural Structure Latrine Drain Water source Water treated Water treatment Cooking fuel Total monthly household expenditure Adivasis Christian ST Dalits OBC (non-muslim) OBC (muslim) Muslim (non-OBC) Constant Observations R-squared

7.4*** (4.80) -1.2 (0.83) -4.3** (2.49) 4.9** (2.51) -2.9 (1.43) 1.5 (0.71) 0.1 (0.06) 1.9 (1.15) -2.2 (1.01) 11.8*** (4.25) 5.0** (2.17) 0.0004* (1.67) -4.9* (1.71) -3.9 (0.94) -7.1*** (3.00) -2.5 (1.29) -8.6** (2.50) -6.1** (2.03) 43.7*** (13.05) 1624 0.08

3

Deaths in Forward States -

Deaths in Backward States -

-1.9 (0.94) -5.7** (2.30) 2.2 (0.83) -0.7 (0.19) 2.5 (0.94) -3.2 (1.10) -2.5 (1.13) -2.9 (1.06) 10.9*** (3.12) 4.6 (1.55) 0.0004

-0.5 (0.25) -2.7 (1.15) 7.8*** (2.63) -3.7 (1.50) 0.5 (0.14) 4.2 (1.25) 6.2** (2.46) -2.3 (0.65) 12.6*** (2.79) 6.2* (1.75) 0.0005

(1.00) -2.9 (0.59) 0.0 (.) -2.1 (0.64) -1.8 (0.69) -5.3 (1.07) -6.8* (1.66) 53.6*** (10.67) 696 0.06

(1.31) -6.7* (1.78) -6.5 (1.39) -11.5*** (3.32) -3.0 (1.04) -11.9** (2.46) -6.1 (1.41) 41.3*** (8.82) 928 0.07

Notes to Table 3 1. Absolute value of t statistics in parentheses. 2. * significant at 10%; ** significant at 5%; *** significant at 1%. 3. “Forward states”: Andhra, Chandigarh, Dadra and Nagar Haveli, Daman and Diu, Delhi, Goa, Gujarat Haryana, Himachal, Karnataka, Kerala, Maharashtra, Punjab, Tamil Nadu, West Bengal. The remaining states and union territories were classified as “backward”. 4. Structure=1 if housing type was pucca, or semi-pucca, or “serviceable” kutcha (i.e. good); 0 otherwise. 5. Latrine=1 if the latrines were flushing toilets or emptied into a sceptic tank; 0 otherwise. 6. Drain=1 if drains were underground or were covered pucca; 0, otherwise. 7. Water source=1 if if the source of drinking water was from a tap; 0 otherwise. 8. Water treated=1 if drinking water treated; 0 otherwise. 9. Water treated=1 if the nature of treatment was boiling, filtering, or ultraviolet/resin/reverse osmosis; 0 otherwise. 10. Cooking fuel=1 if the cooking fuel was gas, gobar gas, kerosene, or electricity; 0 otherwise.

4

Table 4: Own Perception of State of Health of Persons 60 years and above, by Social Group Excellent/ Good/Fair Poor Total very good Health health health Adivasis 143 1,525 428 2,096 6.8 72.8 20.4 100.00 Christian ST 76 534 115 725 10.5 73.7 15.9 100.00 Dalits 220 3,440 1,423 5,083 4.3 67.7 28.0 100.00 OBC (non-muslim) 529 7,848 2,746 11,123 4.8 70.6 24.7 100.00 OBC (muslim) 73 819 409 1,301 5.6 63.0 31.4 100.00 Hindus FC 629 6,867 2,179 9,675 6.5 71.0 22.5 100.00 Muslims (non-OBC) 73 1,315 628 2,016 3.6 65.2 31.2 100.00 Christians (non-ST) 29 328 143 500 5.8 65.6 28.6 100.00 Sikhs 32 295 78 405 7.9 72.8 19.3 100.00 Other religions 23 158 50 231 10.0 68.4 21.6 100.00 Total 1,827 23,129 8,199 33,155 5.5 69.8 24.7 100.00

5

Table 5: Marginal Probabilities from the ordered logit model of own perception of state of health: persons 60 years and above Forward State Age Age squared Female Low education Medium education High education Rural Structure Latrine Drain Water source Water treated Water treatment Cooking fuel Lowest quartile of monthly expenditure Second quartile of monthly expenditure Third quartile of monthly expenditure Adivasis Christian ST Dalits OBC (non-muslim) OBC (muslim) Muslim (non-OBC) Christian (nontribal) Sikh Other religions

Poor Health

Good/Fair Health

-0.011** (2.33) 0.052*** (11.95) -0.0003*** (8.95) 0.043*** (9.22) -0.016*** (2.82) -0.042*** (5.76) -0.074*** (8.24) -0.009 (1.44) -0.032*** (4.38) -0.010 (1.59) -0.002 (0.26) -0.031*** (6.20) -0.043*** (5.88) 0.054*** (5.88) -0.034*** (5.12) 0.044***

0.008** (2.32) -0.039*** (11.80) 0.0002*** (8.88) -0.032*** (9.17) 0.012*** (2.86) 0.030*** (6.11) 0.048*** (10.06) 0.007 (1.46) 0.025*** (4.26) 0.008 (1.59) 0.001 (0.26) 0.023*** (6.22) 0.032*** (6.81) -0.042*** (5.67) 0.025*** (5.20) -0.034***

Excellent/Very Good Health 0.003** (2.33) -0.013*** (11.75) 0.0001*** (8.88) -0.011*** (9.09) 0.004*** (2.71) 0.012*** (4.98) 0.025*** (6.00) 0.002 (1.46) 0.007*** (4.79) 0.003 (1.57) 0.0004 (0.26) 0.008*** (6.06) 0.012*** (6.12) -0.012*** (6.65) 0.009*** (4.85) -0.010***

(6.20) 0.037***

(6.03) -0.029***

(6.73) -0.008***

(4.77) 0.022***

(4.63) -0.017***

(5.25) -0.005***

(3.58) -0.050*** (5.81) -0.083*** (7.07) 0.026** (3.37) 0.001 (0.29) 0.055*** (4.19) 0.081*** (7.23) 0.035*

(3.53) 0.035*** (6.41) 0.052*** (9.79) -0.020*** (3.30) -0.001 (0.29) -0.043*** (4.01) -0.065*** (6.84) -0.027*

(3.73) 0.015*** (4.74) 0.031*** (4.75) -0.006*** (3.60) -0.0004 (0.29) -0.011*** (5.29) -0.016*** (9.16) -0.008**

(1.79) -0.026 (1.59) -0.021 (0.82) 33130

(1.74) 0.018* (1.66) 0.015 (0.85) 33130

(2.05) 0.007 (1.43) 0.006 (0.75) 33130

Observations See notes to Table 3 Low education: literate without schooling, below primary, primary. Medium education: middle or secondary school. High education: higher secondary or more.

6

Table 6: Ailments of persons 60 years and above who regarded their state of health as “poor”, by social group Soc Grp Ailm GASTR

Adv

ST, CH

Dalit

OBC nonMusl

OBC Muslim

Hindu FC

Musl non OBC

CH, nonST

Sikh

Oth

TOT

13 5.04

13 16.25

48 5.13

96 5.43

14 4.70

91 5.77

55 12.39

4 3.45

1 1.85

0 0.00

335 6.02

CARD

23 8.91

3 3.75

85 9.09

223 12.62

49 16.44

298 18.88

71 15.99

38 32.76

10 18.52

7 18.92

807 14.50

RESP

28 10.85

10 12.50

130 13.90

233 13.19

46 15.44

187 11.85

60 13.51

7 6.03

8 14.81

2 5.41

711 12.77

JOINT

32 12.40

8 10.00

117 12.51

248 14.04

25 8.39

186 11.79

49 11.04

22 18.97

10 18.52

5 13.51

702 12.61

KIDNY

5 1.94

0 0.00

20 2.14

30 1.70

3 1.01

44 2.79

6 1.35

2 1.72

3 5.56

1 2.70

114 2.05

NEURO

10 3.88

2 2.50

43 4.60

71 4.02

16 5.37

86 5.45

23 5.18

3 2.59

1 1.85

2 5.41

257 4.62

EYES

26 10.08

3 3.75

121 12.94

154 8.72

22 7.38

103 6.53

41 9.23

11 9.48

1 1.85

3 8.11

485 8.71

DIABT

4 1.55

2 2.50

13 1.39

65 3.68

13 4.36

101 6.40

19 4.28

9 7.76

2 3.70

3 8.11

231 4.15

FEVER

10 3.88

6 7.50

35 3.74

53 3.00

6 2.01

27 1.71

12 2.70

1 0.86

1 1.85

0 0.00

151 2.71

DISAB

63 24.42

26 32.50

153 16.36

282 15.96

53 17.79

183 11.60

41 9.23

6 5.17

8 14.81

7 18.92

822 14.77

ACC

4 1.55

0 0.00

18 1.93

40 2.26

7 2.35

46 2.92

5 1.13

0 0.00

2 3.70

0 0.00

122 2.19

CANC

3 1.16

1 1.25

17 1.82

33 1.87

4 1.34

31 1.96

4 0.90

3 2.59

0 0.00

1 2.70

97 1.74

OTHER

37 14.34

6 7.50

135 14.44

239 13.53

40 13.42

195 12.36

58 13.06

10 8.62

7 12.96

6 16.22

733 13.17

TOT

258 100

80 100

935 100

1,767 100

298 100

1,578 100

444 100

116 100

54 100

37 100

5,567 100

Definition of Ailments: 1. Gastro-intestinal problems (GASTR) 2. Cardiovascular disease (CARD) 3. Respiratory problems (RESP) 4. Disorders of the joints (JOINT) 5. Diseases of the kidney or urinary system (KIDNY) 6. Neurological disorders (NEURO) 7. Eye disorders (EYES) 8. Diabetes (DIABT) 9. Fever-related illness (FEVER) 10. Disabilities (DISAB) 11. Accidents/injuries/burns (ACC) 12. Cancer (CANC) 13. Other ailments (OTHER)

Source: NSS 60th Round, Health File

7

Table 7: Proportion of Persons 60 years, and who regarded their state of health as “poor”, taking treatment for reported ailment, by social group Not taking Taking Treatment Total treatment Adivasis 98 157 255 38.43 61.57 100.00 Christian ST 33 42 75 44.00 56.00 100.00 Dalits 302 626 928 32.54 67.46 100.00 OBC (non470 1,273 1,743 Muslim) 26.97 73.03 100.00 OBC (Muslim) 62 233 295 21.02 78.98 100.00 Hindus FC 239 1,326 1,565 15.27 84.73 100.00 Muslims (non118 319 437 OBC) 27.00 73.00 100.00 Christians (non13 102 115 ST) 11.30 88.70 100.00 Sikhs 10 44 54 18.52 81.48 100.00 Other religions 4 32 36 11.11 88.89 100.00 Total 1,349 4,154 5,503 24.51 75.49 100.00 Source: NSS 60th Round, Health File

8

Table 8: Marginal probabilities from the logit model of treatment received for ailments: persons, 60 years and above, who regarded their state of health as “poor”

Other religions

Marginal Probability of Receiving Treatment for Reported Ailment 0.110*** (9.28) 0.013 (1.01) 0.066*** (4.67) 0.108*** (5.82) 0.094*** (3.11) 0.083*** (2.85) -0.21 (0.98) -0.016 (1.22) 0.049*** (3.23) 0.035** (2.08) 0.004 (0.15) 0.051*** (3.93) 0.055*** (2.92) -0.009 (0.69) -0.161*** (7.84) -0.115*** (5.19) -0.046*** (2.64) -0.106*** (3.15) -0.225*** (3.59) -0.090*** (4.20) -0.057*** (3.34) -0.007 (0.24) -0.077** (2.82) 0.017 (0.34) -0.059 (0.97) 0.068

Pseudo-R-squared Observations

0.0793 5484

Forward State Female Low education Medium education High education Living son(s) Living daughter(s) Rural Economically independent Economically partially dependent Living alone Living with spouse Confined to bed Confined to home Lowest quartile of monthly expenditure Second quartile of monthly expenditure Third quartile of monthly expenditure Adivasis Christian ST Dalits OBC (non-muslim) OBC (muslim) Muslim (non-OBC) Christian (non-tribal) Sikh

Dependent variable = 1 if treatment received for reported ailment, =0, if not received Absolute value of z values in parentheses * significant at 10%; ** significant at 5%; *** significant at 1%

9

Table 9: The Predictive Performance of the Logit Model of the probability of taking treatment for ailment Taking Not taking Total treatment treatment (M) (~M) Predicted as taking 4027 1211 5238 treatment (+) P(M|+) = 76.9% Predicted as not 115 131 246 taking treatment (-) P(~M|-)=53.3% Total 4142 1342 5484 P(+|M) = 90.2% P(-|~M)=2.8% Correctly classified = 75.8% Persons, 60 years and above, who regarded their state of health as “poor”

10

Table 10 Proportion of ever married women who did not receive pre- and post-natal care Pre-Natal Post-Natal Care Care Adivasi 30.5 43.7 Christian ST 37.9 44.1 Dalits 26.2 36.5 OBC (non-Muslim) 22.7 31.3 OBC (Muslim) 32.7 36.4 Hindu (FC) 14.7 26.7 Muslim (non-OBC) 26.1 34.5 Christian (non-ST) 1.5 10.4 Sikh 18.3 31.8 Other Religion 14.3 41.0 Total 23.5 33.2 th Source: NSS 60 Round, Health File

11

Table 11: Marginal probabilities from the logit model of pre- and pot-natal care Forward State Age Low education Medium education High education Rural Labourer Lowest quartile of monthly expenditure Second quartile of monthly expenditure Third quartile of monthly expenditure Adivasis Christian ST Dalits OBC (non-muslim) OBC (muslim) Muslim (non-OBC) Christian (non-tribal) Sikh Other religions Pseudo-R-squared Observations

12

Pre-natal care 0.153*** (17.8) -0.002*** (3.56) 0.090*** (10.5) 0.140*** (16.0) 0.157*** (16.9) -0.028** (2.85) -0.011 (0.64) -0.033**

Post-natal care 0.019 (1.53) -0.001 (0.59) 0.040** (2.73) 0.110*** (7.64) 0.140*** (7.81) -0.047*** (3.55) -0.020 (0.85) -0.010

(2.33) -0.045***

(0.52) -0.013

(3.33) -0.020*

(0.73) -0.017

(1.81) -0.025 (1.38) -0.115*** (4.19) -0.019 (1.25) -0.003 (0.21) -0.088*** (3.62) -0.043** (2.16) 0.163*** (4.46) -0.110* (1.86) 0.039 (0.67) 0.113 9,696

(1.14) -0.082*** (3.14) -0.123*** (3.09) -0.030 (1.51) 0.004 (0.25) -0.041 (1.40) -0.029 (1.15) 0.173 (2.78) 0.005 (0.09) -0.163* (1.96) 0.028 6,874